{2025.06} rocBLAS/4.4.0-rocm-compilers-19.0.0-ROCm-6.4.1 - #1549
Conversation
|
By now, easybuilders/easybuild-easyconfigs#25576 is part of the EasyBuild 5.3.1. Please move this to an |
Build with EasyBuild 5.3.1 (rocBLAS-4.4.0 is included in easybuild-easyconfigs-v5.3.1), so no from-commit option is needed.
b17fe36 to
732635e
Compare
|
@casparvl I've moved rocBLAS to the |
|
bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx908 |
|
Looks like some non-ascii character is causing failures in delivering this event to the bot. I suspect it was a non-standard hyphen in the first message, replaced it by a regular hyphen. Let's try again. bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx908 |
|
There was also still a bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx908 |
|
New job on instance
|
|
bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx90a+amd/gfx942+amd/gfx1030+amd/gfx1100+amd/gfx1101+amd/gfx1200+amd/gfx1201 |
|
New job on instance
|
|
bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen3 for:arch=x86_64/amd/zen3,accel=amd/gfx908+amd/gfx90a+amd/gfx942+amd/gfx1030+amd/gfx1100+amd/gfx1101+amd/gfx1200+amd/gfx1201 |
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx90a |
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
New job on instance
|
|
Trying to debug and solve the build issue in EESSI/software-layer-scripts#281, but it's not clear to me what's going on. Some builds just randomly stall and get aborted/requeued by Slurm. Initially it looked like an OOM issue, but that doesn't seem to be the case. |
|
bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx942 |
|
New job on instance
|
|
bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=skylake_avx512 for:arch=x86_64/intel/skylake_avx512,accel=amd/gfx942 |
|
New job on instance
|
|
New job on instance
|
|
Sapphirerapids should (hopefully) be almost done, and then this is the very last one... bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=zen2 for:arch=x86_64/amd/zen2,accel=amd/gfx942 |
|
New job on instance
|
|
bot:status last_build |
|
This is the status of all the
|
8 similar comments
|
This is the status of all the
|
|
This is the status of all the
|
|
This is the status of all the
|
|
This is the status of all the
|
|
This is the status of all the
|
|
This is the status of all the
|
|
This is the status of all the
|
|
This is the status of all the
|
|
Cascadelake and icelake have tarballs for all 8 gfx targets combined. zen{2,3,4,5}, generic, haswell, sapphirerapids, skylake have individual tarballs for the 8 gfx targets. So, this is finally ready! We do need to merge EESSI/software-layer-scripts#302 and easybuilders/easybuild-easyconfigs#26846 first. |
|
Thought I had done the last couple of builds with 8 cores for rocBLAS, but it was accidentally still set to 1. Let me check if it also works with 8 using the updated easyconfig. bot: build repo:eessi.io-2025.06-software instance:eessi-bot-mc-aws on:arch=skylake_avx512 for:arch=x86_64/intel/skylake_avx512,accel=amd/gfx942 |
|
New job on instance
|
|
@bedroge I've merged the easyblock PR, the scripts PR is still in draft just let me know when it is ready |
Which easyblock PR is that? (couldn't find one referenced in the bot, but I only looked very briefly) |
I guess he meant easyconfig: easybuilders/easybuild-easyconfigs#26846. |
bedroge
left a comment
There was a problem hiding this comment.
Lgtm. Not all builds were done with the exact same hook and exact same easyconfig, but the only difference would be the number of cores used for hipBLASLt/rocBLAS and the value for Tensile_CPU_THREADS (see easybuilders/easybuild-easyconfigs#26846). That doesn't affect the final binaries in any way, so no need to do these very expensive builds again imho.
|
Staging channel on Slack shows that the combined tarballs (containing builds for all 8 gfx targets) have been ingested for cascadelake and icelake, and 8 individual tarballs have been ingested for all other x86_64 targets (zen{2,3,4,5}, generic, haswell, sapphirerapids, skylake_avx512). So, this is finally done! 🥳 |
Add
rocBLAS-4.4.0-rocm-compilers-19.0.0-ROCm-6.4.1to theaccel/amdeasystack for
2025.06(eb-5.3.0-2025a).rocBLAS is the ROCm BLAS library and the natural next component to sit on top of
the
rocm-compilers/ ROCm 6.4.1 stack already prioritized here (#1499). Itunlocks GPU-accelerated dense linear algebra on AMD GPUs for the toolchain -
e.g. transparently offloading
DGEMMfrom FlexiBLAS-linked apps (HPL, Octave,NumPy/SciPy, ...) via a rocBLAS hook.
The easyconfig comes from the same upstream PR as the sibling
rocm-compilersentry (easybuilders/easybuild-easyconfigs#25576), so it is pinned with
from-commitin the same style. Its build/runtime dependencies (Tensile4.43.0, hipBLASLt 0.12.1, rocTracer 4.1.0, HIP 6.4.1) are all present on the
same commit.
easybuild/easyconfigs/r/rocBLAS/rocBLAS-4.4.0-rocm-compilers-19.0.0-ROCm-6.4.1.ebrocm-compilers/19.0.0-ROCm-6.4.1from-commit: 04fa96aa0041aa74e2a851ba1e22af411a800d75(last commit touching the easyconfig on #25576)